中圖分類號：TP391
文獻標識碼：A
DOI： 10.19358/j.issn.2096-5133.2020.02.013
引用格式：俞旭峰，王澎，郭威，等.基于時空聚集的網(wǎng)貸反欺詐建模與研究［J］.信息技術與網(wǎng)絡安全，2020,39（2）：69-74.

Anti-fraud modeling and research of online loans based on time and space aggregation

Abstract： The identification of sudden gang fraud has become an urgent problem in the online loan business.In the case of less feature dimensions,this paper proposes an anti-fraud model of online loans based on spatiotemporal aggregation.Firstly,based on the users′ location information and the time of applying for the loan,a clustering indicator suitable for the online loan business，K-N nearest neighbor index is designed;Then,the K-N nearest neighbor index of different time observation windows is used to extract embedding features from seq2seq (sequence to sequence) model based on LSTM (Long Short-Term Memory);Finally,the LightGBM model is used to predict the probability of fraud.The experimental results show that the proposed indicator can capture bad debts more effectively.Compared with only using the basic features,the KS value and AUC of the prediction result are better improved.

Key words : data mining;financial fraud identification;spatiotemporal data analysis;neighbor index;LSTM